Docdex
Index and search documentation inside each repo so your coding agent answers from your real docs instead of guessing APIs.
Overview
docdex is a MCP server for the Build phase that indexes per-repo documentation and serves full-text search to coding agents.
What is this MCP server?
- Fast per-repository documentation indexing
- Full-text search across indexed doc corpora in the repo
- MCP server (docdex v0.2.9) over stdio via npm
- Purpose-built for codebases rather than generic web search
- Keeps retrieval scoped to the project the agent is working in
- Server version 0.2.9 on npm identifier docdex
- Transport: stdio
- Repository: github.com/bekirdag/docdex
Community signal: 20 GitHub stars.
What problem does it solve?
Agents waste tokens and ship wrong integrations because scattered repo docs are not searchable in one place during implementation.
Who is it for?
Solo builders with non-trivial READMEs, API notes, or internal docs who want MCP-native search without standing up a separate doc SaaS.
Skip if: Greenfield repos with no documentation to index, or teams that already rely solely on external wikis with a different MCP integration.
What do I get? / Deliverables
After install, your agent can query indexed documentation in the active repo and cite the sections that actually exist in your tree.
- Indexed documentation corpus for the active repo
- MCP-accessible full-text search over that index
- Faster grounded answers when agents implement against your docs
Recommended MCP Servers
Journey fit
How it compares
Repo-scoped doc search MCP, not a general RAG pipeline skill or marketplace catalog.
Common Questions / FAQ
Who is docdex for?
Developers using Claude Code, Cursor, or Codex who maintain real documentation in the repo and want agents to search it via MCP.
When should I use docdex?
Use it in Build while implementing features, wiring integrations, or updating docs-heavy modules where fast full-text lookup beats manual file grep.
How do I add docdex to my agent?
Install npm package docdex (v0.2.9), add it as a stdio MCP server in your client configuration, and point indexing at the repository root per the bekirdag/docdex README.